Event Sourcing

When too many people try to update the same one record at once, retries pile up. Fixes usually mean either spreading that one record's updates across several smaller pieces, changing the kind of update so order stops mattering, or accepting some updates outside the strict record and reconciling later — each trades away some strict consistency or simplicity for more speed.

If an event type has been changed a dozen times, the oldest events might need to run through eleven small translation steps every time they're replayed, which is slow and hard to maintain. Teams fix this by periodically rewriting old streams into the current shape once, so future reads skip the whole chain.

Build the new, corrected version in a completely separate table while the old one keeps serving traffic, and only switch customers over to the new one once it's fully caught up and checked - like renovating a store in a new building instead of closing the old one down for the day.